Hi everyone! I’m Lucas, the solo developer behind ZeroLeaks. Over the past year I’ve been researching prompt injection and system prompt exposure in AI agents. While testing real agent workflows I kept seeing the same pattern: once agents can read external content and call tools, prompt injection becomes a real security problem. That’s why I built ZeroLeaks. It’s a tool that automatically runs security scans against AI agents to detect vulnerabilities like prompt injection and tool misuse before they reach production. Some things ZeroLeaks can do: • test agents against prompt injection attempts • detect system prompt exposure • simulate adversarial instructions • analyze tool misuse risks The project is open source, and I’ve been building it mostly solo while documenting my findings publicly. Some of those scans have already surfaced interesting vulnerabilities in real AI agents. If you’re building with AI agents or LLM workflows, I’d love to hear: – what security concerns you have today – what kinds of attacks worry you the most – what you’d want a tool like this to detect Happy to answer any questions and super grateful for the feedback 🙏

ZeroLeaks

Security testing for AI agents

Security testing for AI agents